*fix bug in AMX-BF16 optimizations of class SynetConvolution32fBf16Nh…

…wcGemm.
ermig1979 · Mar 13, 2024 · 636728d · 636728d
1 parent 5ca9f05
commit 636728d
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 2 deletions.
diff --git a/src/Simd/SimdAmxBf16SynetConvolution32fBf16NhwcGemm.cpp b/src/Simd/SimdAmxBf16SynetConvolution32fBf16NhwcGemm.cpp
@@ -100,7 +100,7 @@ namespace Simd
             else
             {
                 //SIMD_PERF_BEG("solid");
-                dst += b * p.dstH * p.dstW * p.dstC;
+                dst += b * p.dstH * p.dstW * a.bufK;
                 for (size_t n = (yEnd - yBeg) * p.srcW * p.srcC, i = 0; i < n; i += 32)
                     Float32ToBFloat16<false, false>(src + i, dst + i, srcMask, dstMask);
             }

diff --git a/src/Test/TestSynetConvolution32f.cpp b/src/Test/TestSynetConvolution32f.cpp
@@ -651,12 +651,14 @@ namespace Test
         //result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 48, 160, 160, 32, _1, _1, _1, _0, _0, 1, a, t), c, f1, f2);
         //result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 32, 80, 80, 32, _3, _1, _1, _1, _1, 1, a, t), c, f1, f2);
         //result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 96, 80, 80, 64, _1, _1, _1, _0, _0, 1, a, t), c, f1, f2);
+        result = result && SynetConvolution32fForwardAutoTest(eps, Param(2, 256, 1, 1, 512, _1, _1, _1, _0, _0, 1, a, t), c, f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 1664, 40, 40, 512, _1, _1, _1, _0, _0, 1, a, t), c, f1, f2);
         result = result && SynetConvolution32fForwardAutoTest(eps, Param(2, 192, 5, 5, 256, _3, _1, _1, _0, _0, 1, a, t), c, f1, f2);
 #endif
 #else
         //result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 16, 160, 160, 16, _3, _1, _1, _1, _1, 1, a, t), c, f1, f2);
-        result = result && SynetConvolution32fForwardAutoTest(eps, Param(2, 192, 5, 5, 256, _3, _1, _1, _0, _0, 1, a, t), c, f1, f2);
+        //result = result && SynetConvolution32fForwardAutoTest(eps, Param(2, 192, 5, 5, 256, _3, _1, _1, _0, _0, 1, a, t), c, f1, f2);
+        result = result && SynetConvolution32fForwardAutoTest(eps, Param(2, 256, 1, 1, 512, _1, _1, _1, _0, _0, 1, a, t), c, f1, f2);
         //result = result && SynetConvolution32fForwardAutoTest(eps, Param(1, 3, 1, 8, 8, _1, _1, _1, _0, _0, 1, aId, tT), bf16, f1, f2);